权重规范$ \ | w \ | $和保证金$ \ gamma $通过归一化的保证金$ \ gamma/\ | w \ | $参与学习理论。由于标准神经净优化器不能控制归一化的边缘，因此很难测试该数量是否与概括有关。本文设计了一系列实验研究，这些研究明确控制了归一化的边缘，从而解决了两个核心问题。首先：归一化的边缘是否总是对概括产生因果影响？本文发现，在归一化的边缘似乎与概括没有关系的情况下，可以与Bartlett等人的理论背道而驰。（2017）。第二：标准化边缘是否对概括有因果影响？该论文发现是的 - 在标准培训设置中，测试性能紧密跟踪了标准化的边距。该论文将高斯流程模型表示为这种行为的有前途的解释。

Training large neural networks requires distributing learning across multiple workers, where the cost of communicating gradients can be a significant bottleneck. SIGNSGD alleviates this problem by transmitting just the sign of each minibatch stochastic gradient. We prove that it can get the best of both worlds: compressed gradients and SGD-level convergence rate. The relative 1 / 2 geometry of gradients, noise and curvature informs whether SIGNSGD or SGD is theoretically better suited to a particular problem. On the practical side we find that the momentum counterpart of SIGNSGD is able to match the accuracy and convergence speed of ADAM on deep Imagenet models. We extend our theory to the distributed setting, where the parameter server uses majority vote to aggregate gradient signs from each worker enabling 1-bit compression of worker-server communication in both directions. Using a theorem by Gauss (1823) we prove that majority vote can achieve the same reduction in variance as full precision distributed SGD. Thus, there is great promise for sign-based optimisation schemes to achieve fast communication and fast convergence. Code to reproduce experiments is to be found at https://github.com/jxbz/signSGD.

Many real-world applications of language models (LMs), such as code autocomplete and writing assistance, involve human-LM interaction, but the main LM benchmarks are non-interactive, where a system produces output without human intervention. To evaluate human-LM interaction, we develop a framework, Human-AI Language-based Interaction Evaluation (H-LINE), that expands non-interactive evaluation along three dimensions, capturing (i) the interactive process, not only the final output; (ii) the first-person subjective experience, not just a third-party assessment; and (iii) notions of preference beyond quality. We then design five tasks ranging from goal-oriented to open-ended to capture different forms of interaction. On four state-of-the-art LMs (three variants of OpenAI's GPT-3 and AI21's J1-Jumbo), we find that non-interactive performance does not always result in better human-LM interaction and that first-person and third-party metrics can diverge, suggesting the importance of examining the nuances of human-LM interaction.

Generating a chain of thought (CoT) can increase large language model (LLM) performance on a wide range of tasks. Zero-shot CoT evaluations, however, have been conducted primarily on logical tasks (e.g. arithmetic, commonsense QA). In this paper, we perform a controlled evaluation of zero-shot CoT across two sensitive domains: harmful questions and stereotype benchmarks. We find that using zero-shot CoT reasoning in a prompt can significantly increase a model's likelihood to produce undesirable output. Without future advances in alignment or explicit mitigation instructions, zero-shot CoT should be avoided on tasks where models can make inferences about marginalized groups or harmful topics.

Prior work has identified a resilient phenomenon that threatens the performance of human-AI decision-making teams: overreliance, when people agree with an AI, even when it is incorrect. Surprisingly, overreliance does not reduce when the AI produces explanations for its predictions, compared to only providing predictions. Some have argued that overreliance results from cognitive biases or uncalibrated trust, attributing overreliance to an inevitability of human cognition. By contrast, our paper argues that people strategically choose whether or not to engage with an AI explanation, demonstrating empirically that there are scenarios where AI explanations reduce overreliance. To achieve this, we formalize this strategic choice in a cost-benefit framework, where the costs and benefits of engaging with the task are weighed against the costs and benefits of relying on the AI. We manipulate the costs and benefits in a maze task, where participants collaborate with a simulated AI to find the exit of a maze. Through 5 studies (N = 731), we find that costs such as task difficulty (Study 1), explanation difficulty (Study 2, 3), and benefits such as monetary compensation (Study 4) affect overreliance. Finally, Study 5 adapts the Cognitive Effort Discounting paradigm to quantify the utility of different explanations, providing further support for our framework. Our results suggest that some of the null effects found in literature could be due in part to the explanation not sufficiently reducing the costs of verifying the AI's prediction.

Hyperspectral Imaging (HSI) provides detailed spectral information and has been utilised in many real-world applications. This work introduces an HSI dataset of building facades in a light industry environment with the aim of classifying different building materials in a scene. The dataset is called the Light Industrial Building HSI (LIB-HSI) dataset. This dataset consists of nine categories and 44 classes. In this study, we investigated deep learning based semantic segmentation algorithms on RGB and hyperspectral images to classify various building materials, such as timber, brick and concrete.

Strategic test allocation plays a major role in the control of both emerging and existing pandemics (e.g., COVID-19, HIV). Widespread testing supports effective epidemic control by (1) reducing transmission via identifying cases, and (2) tracking outbreak dynamics to inform targeted interventions. However, infectious disease surveillance presents unique statistical challenges. For instance, the true outcome of interest - one's positive infectious status, is often a latent variable. In addition, presence of both network and temporal dependence reduces the data to a single observation. As testing entire populations regularly is neither efficient nor feasible, standard approaches to testing recommend simple rule-based testing strategies (e.g., symptom based, contact tracing), without taking into account individual risk. In this work, we study an adaptive sequential design involving n individuals over a period of {\tau} time-steps, which allows for unspecified dependence among individuals and across time. Our causal target parameter is the mean latent outcome we would have obtained after one time-step, if, starting at time t given the observed past, we had carried out a stochastic intervention that maximizes the outcome under a resource constraint. We propose an Online Super Learner for adaptive sequential surveillance that learns the optimal choice of tests strategies over time while adapting to the current state of the outbreak. Relying on a series of working models, the proposed method learns across samples, through time, or both: based on the underlying (unknown) structure in the data. We present an identification result for the latent outcome in terms of the observed data, and demonstrate the superior performance of the proposed strategy in a simulation modeling a residential university environment during the COVID-19 pandemic.

It is crucial to choose the appropriate scale in order to build an effective and informational representation of a complex system. Scientists carefully choose the scales for their experiments to extract the variables that describe the causalities in the system. They found that the coarse scale(macro) is sometimes more causal and informative than the numerous-parameter observations(micro). The phenomenon that the causality emerges by coarse-graining is called Causal Emergence(CE). Based on information theory, a number of recent works quantitatively showed that CE indeed happens while coarse-graining a micro model to the macro. However, the existing works have not discussed the question of why and when the CE happens. We quantitatively analyze the redistribution of uncertainties for coarse-graining and suggest that the redistribution of uncertainties is the cause of causal emergence. We further analyze the thresholds that determine if CE happens or not. From the regularity of the transition probability matrix(TPM) of discrete systems, the mathematical expressions of the model properties are derived. The values of thresholds for different operations are computed. The results provide the critical and specific conditions of CE as helpful suggestions for choosing the proper coarse-graining operation. The results also provided a new way to better understand the nature of causality and causal emergence.

Artificial intelligence (AI) has enormous potential to improve Air Force pilot training by providing actionable feedback to pilot trainees on the quality of their maneuvers and enabling instructor-less flying familiarization for early-stage trainees in low-cost simulators. Historically, AI challenges consisting of data, problem descriptions, and example code have been critical to fueling AI breakthroughs. The Department of the Air Force-Massachusetts Institute of Technology AI Accelerator (DAF-MIT AI Accelerator) developed such an AI challenge using real-world Air Force flight simulator data. The Maneuver ID challenge assembled thousands of virtual reality simulator flight recordings collected by actual Air Force student pilots at Pilot Training Next (PTN). This dataset has been publicly released at Maneuver-ID.mit.edu and represents the first of its kind public release of USAF flight training data. Using this dataset, we have applied a variety of AI methods to separate "good" vs "bad" simulator data and categorize and characterize maneuvers. These data, algorithms, and software are being released as baselines of model performance for others to build upon to enable the AI ecosystem for flight simulator training.

Real-time air pollution monitoring is a valuable tool for public health and environmental surveillance. In recent years, there has been a dramatic increase in air pollution forecasting and monitoring research using artificial neural networks (ANNs). Most of the prior work relied on modeling pollutant concentrations collected from ground-based monitors and meteorological data for long-term forecasting of outdoor ozone, oxides of nitrogen, and PM2.5. Given that traditional, highly sophisticated air quality monitors are expensive and are not universally available, these models cannot adequately serve those not living near pollutant monitoring sites. Furthermore, because prior models were built on physical measurement data collected from sensors, they may not be suitable for predicting public health effects experienced from pollution exposure. This study aims to develop and validate models to nowcast the observed pollution levels using Web search data, which is publicly available in near real-time from major search engines. We developed novel machine learning-based models using both traditional supervised classification methods and state-of-the-art deep learning methods to detect elevated air pollution levels at the US city level, by using generally available meteorological data and aggregate Web-based search volume data derived from Google Trends. We validated the performance of these methods by predicting three critical air pollutants (ozone (O3), nitrogen dioxide (NO2), and fine particulate matter (PM2.5)), across ten major U.S. metropolitan statistical areas (MSAs) in 2017 and 2018.